Start
Dec 13, 2007 - 15:30
End
Dec 13, 2007 - 16:30
Venue
CDEEP Seminar Hall KRESIT (3rd floor)
Event Type
Speaker
Prof. Lorenz T. Biegler
Title
Trust Region Algorithms for Parameter Estimation
Model discrimination parameter estimation and statistical inference require the formulation and efficient solution of nonlinear programming (NLP) problems. Frequently these problems are ill-posed due to over-parameterized models or data with incomplete information. These characteristics lead to failure of many popular NLP codes. This talk presents some basic parameter estimation methods that can also handle structured features of ill-posed parameter estimation problems. These are treated with composite step trust region methods. A key point for both methods is that they have guaranteed convergence properties and allow further analysis for model building and discrimination. Case studies based on parameter estimation in polymerization reactors are presented that illustrate these concepts.